Diagnose Aurora Databases Directly from Your Agent

Where this fits

You're building an agent that handles database incidents for Aurora MySQL or PostgreSQL. CloudWatch metrics and RDS API calls tell you CPU, storage, and connections—but they don't show you what's happening inside the buffer pool, which queries are holding locks, or whether replication lag is caused by a specific index. Your agent gets stuck because it lacks a safe way to run diagnostic queries. The database-rds-devops skill fills that gap by exposing predefined, read-only health checks through the RDS Data API, so your agent can inspect database internals without credentials or manual SQL.

Why agents benefit

• Runs read-only diagnostics safely – All queries are predefined and read-only, so the agent can execute them without risk of accidental writes or schema changes.
• Covers multiple health categories – Buffer pool, connections, locks, replication, storage, performance, and index efficiency are all addressed via 10 MCP tools, giving the agent a structured way to investigate.
• Works across Aurora engines – Supports both Aurora MySQL and PostgreSQL, so a single skill can handle mixed environments.
• Reduces tool sprawl – Instead of wiring custom SQL execution, the agent calls run_full_health_check or run_category_check to get a consolidated view.
• Integrates with existing MCP infrastructure – Uses the rds-aidba MCP server over streamable HTTP with SigV4 auth, so it fits into your agent's existing tool ecosystem.

Practical scenarios

Diagnosing a sudden connection spike

Your agent receives an alert about max connections reached. It calls run_category_check with category connections on the affected cluster. The results show active sessions, wait events, and which user or application is consuming them. The agent can then recommend terminating idle sessions or scaling instance class—backed by evidence.

Investigating replication lag before failover

A replica is falling behind, and your agent must decide if failover is safe. It runs get_cluster_health to confirm cluster status, then executes the replication category queries to check Seconds_Behind_Master and identify long-running transactions blocking apply. The agent can then advise whether to wait or promote the replica.

Proactive index efficiency audit

During a performance review, your agent runs run_full_health_check across all clusters. It flags tables with high index scan ratios or missing primary keys. The agent generates a report and suggests specific indexes to add, using the query output as justification.

Add it to your agent workflow

Install the skill using the OpenAgentSkill CLI:

npx skills add aws/tools-for-devops-agent --skill database-rds-devops

Then configure the rds-aidba MCP server (see the skill's mcp/rds-aidba/ directory for Lambda deployment). Once running, your agent can invoke tools like:

{
  "tool": "run_full_health_check",
  "parameters": {
    "engine": "aurora-postgresql",
    "cluster_identifier": "prod-main"
  }
}

This returns a JSON payload with health metrics across all categories, which your agent can parse and act on.

Compare before adopting

Before adding this skill, evaluate it against alternatives:

• Quality signals – Check if the predefined queries match your database version and workload patterns. Look at the repository's issues and PRs for community feedback.
• Maintenance freshness – The skill is from AWS (aws/tools-for-devops-agent) with 42 stars. Verify the last commit date and whether it tracks Aurora engine updates.
• Alternatives – You might already have an internal SQL runner or use AWS Performance Insights. Compare the depth of diagnostics versus the operational overhead of deploying the MCP server.
• Workflow fit – Does your agent already use MCP? If not, the SigV4 auth and Lambda setup might add complexity. Ensure the skill's read-only scope aligns with your security policies.

Why it is worth tracking

This skill is a strong example of bringing database-level observability into agent workflows—something CloudWatch alone can't provide. The fact that it's authored by AWS and part of a broader devops agent toolkit suggests ongoing investment. With 42 stars and a focused scope, it's early but promising. Track it if you're building agents that need to handle Aurora incidents autonomously, and evaluate it when your current diagnostics rely on manual SQL or incomplete metrics. For a curated view of similar agent skills, check OpenAgentSkill to see how it compares with other database-focused tools.